About

Ousmane Dicko

Lire en français →

I'm the founder of Haki, a self-hosted memory layer for AI agents, and of Dicken AI, Mali's first AI automation agency. I work from Bamako. Every article on this site is signed by me, and this page explains who's behind the byline — and how the writing gets made.

The short version

I started Dicken AI as a teenager, building AI agents for African businesses — customer-service bots that answer on WhatsApp while the owner sleeps. TRT Afrika covered that work: agents handling up to 90% of customer interactions for the businesses that deploy them.

Running those agents in production taught me the problem I now work on full-time. An agent that answers well but remembers nothing is a demo. Real clients change their minds, correct you, come back three weeks later — and every framework I tried either forgot everything, or worse, remembered the wrong version with total confidence. The memory layers on the market wanted our conversation data on their servers, which for most of my clients was a non-starter before we even discussed price.

So we built the thing we couldn't buy: memory that runs on your own Postgres, marks old facts as superseded instead of silently overwriting them, and can show you why the agent said what it said. That's Haki — Apache 2.0, docker compose up, no account required. Whether it holds up to that description is not something you should take from me: haki verify proves the memory loop on your machine in about sixty seconds, and a full reliability report with methodology and per-question output is coming once the calibrated measurement run is complete.

In the press

I also write Inside AI, a LinkedIn newsletter on practical AI for French-speaking professionals, read by 9,000+ subscribers.

How the writing here gets made

Two commitments, because this site asks you to trust its numbers:

Every number is sourced or it doesn't ship. Benchmark figures only appear in the reliability report, with the judge, the baseline, and the per-question output published. Where a competitor is better than us, the page says so — the Haki vs Mem0 comparison opens with the section where Mem0 wins.

AI helps with research and drafts; it doesn't get the byline. I use AI tooling to collect sources and prepare drafts — it would be strange not to, given what we build. Every article is then edited, verified and signed by me, every code snippet is executed against a real Haki instance before publishing, and if something's wrong, it's my name on it. If you find an error, open an issue — corrections get a changelog line, not a silent edit.

Contact

The fastest way to reach me is GitHub or ousmane@dickenai.com. If you're building agents and memory is where they break, I genuinely want to hear the details — that feedback is where Haki's roadmap comes from.